3d Point Errors and Change Detection Accuracy of Unmanned Aerial Vehicle Laser Scanning Data

Abstract

Abstract. Unmanned aerial vehicle laser scanning (ULS) has recently become available for operational mapping and monitoring (e.g. for forestry applications or erosion studies). It combines advantages of terrestrial and airborne laser scanning, but there is still little proof of ULS accuracy. For the detection and monitoring of small-magnitude surfaces changes with multitemporal point clouds, an estimate of the level of detection (LOD) is required. The LOD is a threshold applied on distance measurements to separate real surface change (e.g. due to erosion or deposition by geomorphic processes) from errors. This paper investigates key components of the error budget for two ULS point clouds acquired for erosion monitoring at a grassland site in the Alps. In addition to the registration error and effects of the local surface roughness, we assess the positional uncertainties of each point that result from laser footprint effects, which are a function of the scanning geometry (including range, incidence angle and beam divergence). By removing erroneous points with an increasingly stricter point error criterion, we illustrate that the positional point errors strongly affect the LOD and discuss how this type of error can be mitigated. Moreover, our experimental results with three different surface classes (bare earth and rock, buildings and grassland) show that the level of detection tends to be slightly better for areas with bare earth and rock than for grass-covered areas (due to their roughness). For all these surface types reliable distance measurements are possible with sub-decimetre levels of detection.

Airborne Laser Scanning Change Detection for Quantifying Geomorphological Processes in High Mountain Regions

“…In future, not only more sophisticated approaches for a (semi-)automatic classification of topographic change according to process domains but also improved methods for validation should be developed. In cases where additional data (such as individual flight strips and trajectory information) is available for the point clouds, the LOD could be further improved (i) by a refinement of the strip adjustment, (ii) by a quantification of the positional uncertainties at point level, and (iii) by a filtering of the points with high positional uncertainties (Mayr et al 2020). Furthermore, a correction of laser return intensity values to obtain reflectances would be interesting for derivation of a snow mask (e.g., Höfle et al 2007).…”

Three Dimensional Change Detection Using Point Clouds: A Review

“…Here, we introduce commonly used evaluation metrics. A well-known uncertainty metric for CD quantification is the level of detection (LODetection) [28,[153][154][155]. It is used as a threshold to consider only real changes (where the magnitude distance is superior to the LODetection at a specific confidence interval) for further analysis and interpretation (using a statistical t-test and an assumption of normal distribution of errors).…”